🔍 Read the full analysis: OpenAI Splits Its Robotics Group: Does Runway's Praxis-1 Point To What's Next? on ThorstenMeyerAI.com
Get business pricing on monitors, keyboards and dev gear
- Business-only prices and quantity discounts
- Tax-exempt purchasing
- Multiple users, one account, clear invoices
TL;DR
Fortune reported that OpenAI has disbanded its stand-alone robotics group, moving researchers into its main research organization and applied robotics and hardware teams into Applied. OpenAI said the change reflects progress applying general-purpose models to robot control; Runway’s newly released Praxis-1 offers a contrasting specialist approach, with reported results that point to a trade-off between simulated task success and response speed.
OpenAI has broken up its stand-alone robotics group, moving its researchers into the company’s main research organization and its applied robotics and hardware teams into the Applied organization, according to Fortune. The reorganization, which OpenAI confirmed to the outlet, comes as Runway releases Praxis-1, an open-weight model built specifically for robot control, putting two different approaches to robotics in focus.
Fortune reported November 13 that OpenAI confirmed the staffing change when asked about it. The robotics researchers now report to Chief Research Officer Mark Chen, while applied robotics and hardware teams have moved to the Applied organization, led by President Greg Brockman. At the CoRL robotics conference in Austin this week, Fortune found the booth OpenAI had listed on the floor plan empty. OpenAI has not published its own announcement about the reorganization.
OpenAI’s spokesperson described the move as a sign of commitment, not a retreat. Fortune quoted the company as saying robotics “remains a core part of OpenAI’s long-term vision” and that it remains committed to bringing general-purpose robots into the world. The spokesperson said researchers had earlier expected to build on video models such as Sora, but that frontier models including GPT-6 Astra had since shown they could be applied directly to robotics.
Runway released Praxis-1 two days before Fortune’s report. The company describes it as a 27-billion-parameter “world action model”; its weights and fine-tuning code are available under the Apache 2.0 license, which permits commercial use. The published benchmark results and timing figures below come from the supplied reports and study descriptions; they do not establish how either model will perform across real-world robots and tasks.
Robot control loop · explainer
Who decides the robot’s next move?
Every robot repeats one loop: cameras see, a model decides, the arm moves. The two bets in this story put a different kind of model into the decide step.
Option 1
A general model in the loop
Example: GPT-6 Astra in NVIDIA’s simulation study
18.56 s per model call on average in the direct set-up. Inference was 97.0% of the estimated episode time; the simulation paused while the model was thinking.
Option 2
A specialist robot model in the loop
Example: Runway’s Praxis-1, a world action model with open weights
About 0.55–0.6 s per call on one H100 server, and each call returns 1.6 s of motion. Needs fine-tuning data for each robot.
General model plans the next stepslow, outside the loop
Option 3 · illustrative
Both, in two layers
The idea Runway and the NVIDIA researchers both describe
Runway calls the two approaches “complementary”: an agent plans, a fast policy carries out each step. NVIDIA’s authors say the best way to combine them “remains an open question”.
Sources: NVIDIA Seattle Robotics Lab, “Astranomy” study of GPT-6 Astra, Oct 14, 2026 (simulation, direct set-up); Runway, Praxis-1 technical report, Nov 2026 (one H100 server). Loops and timelines are schematic; the two figures come from different set-ups, not from a head-to-head test.
Graphic: Powered by Thorsten Meyer AI
Speed Versus Success in Robot Control
The split between the approaches is visible in reported results for RoboLab-120, a simulation benchmark of 120 tabletop tasks. In a study by NVIDIA researchers, GPT-6 Astra reached a 74.8% success rate in an agentic setup, compared with 51.0% for the best trained robot policy listed on the benchmark leaderboard at the time, according to the source material. The study also reported a 66.3% result for a direct setup and 66.5% when Astra reviewed another model’s proposed movements.
Higher simulated success does not by itself show that a general model is ready to control physical robots at useful speeds. In the study’s direct setup, each GPT-6 Astra call averaged 18.56 seconds, and the simulation paused while the model processed its response. Praxis-1’s reported score was 46.2% on the same benchmark; Runway said it took about 0.55 to 0.6 seconds per call to produce 1.6 seconds of motion. These figures use different setups and should not be treated as a direct, independent comparison of real-world performance.
Runway also said that, on one of its own robots and across three tasks, GPT-6 Astra took 15.6 times longer than Praxis-1 to finish and cost 51 times more per completed task. That is a company-reported test, not an independent evaluation. Taken together, the available results suggest a practical question for robotics developers: whether a broad model’s capabilities can be paired with faster, more specialized control, rather than choosing one approach outright.
Popular right now
De'Longhi La Specialista Touch Semi-Automatic Espresso Machine with Grinder
Bose QuietComfort Wireless Bluetooth Headphones – Active Over‑Ear Noise Cancelling with Mic, USB‑C Charging, Deep Bass, up to 24 h Playtime,
KitchenAid Artisan 5-Qt Tilt-Head Stand Mixer, KSM150PS, Almond Cream
As an affiliate, we earn on qualifying purchases.
OpenAI’s Changing Robotics Strategy
This is the second time OpenAI has ended a stand-alone robotics team. The company wound down an earlier group in 2020. It rebuilt a robotics team in 2024 and partnered with humanoid-robot maker Figure; Figure said the partnership ended in February 2025 as it pursued its own robot AI, according to the source material.
OpenAI’s current explanation is that general-purpose models have become more useful for robot control than the company expected earlier this year. Its spokesperson connected that progress to the decision to place robotics researchers alongside the teams developing general models. The reorganization does not, on its own, establish that OpenAI has stopped developing robots: the company has said it is working on a humanoid robot, and CEO Sam Altman has targeted a first public demonstration in 2027.
Runway’s release presents the other side of the current debate: a model designed for robot action, released with weights and code that developers can inspect and adapt. The supplied material says Runway calls the specialist and general-purpose approaches “complementary”. That framing is consistent with the reported gap between benchmark success and response time, but whether companies will combine the approaches in deployed systems remains unsettled.
“The two approaches are “complementary.””
— Runway
What the Published Results Cannot Show
It is not clear whether OpenAI will publish a fuller account of the reorganization or specify how the newly assigned teams will work together. Fortune reported that OpenAI confirmed the change in response to questions, while the company had not posted its own announcement as of November 14.
The benchmark evidence is limited. RoboLab-120 is a simulation, and the reported figures come from different model setups; the source material does not establish that the test results predict performance on physical robots. The available information also does not provide an independent replication of Runway’s comparison on its own robot, or enough detail to generalize its timing and cost figures to other tasks and hardware.
OpenAI’s longer-term plans are also not fully specified in the material. The company has said it is developing a humanoid robot and Altman has targeted a 2027 public demonstration, but the robot’s capabilities, schedule, and connection to the reorganized teams remain unclear. It is not yet known whether OpenAI will train specialized control systems alongside its general models or rely more heavily on the latter.
The 2027 Robot Demonstration
The next public milestone cited in the source material is OpenAI’s targeted 2027 humanoid-robot demonstration. The company has not provided further details here about when it will take place or what it will show. Runway’s release also gives researchers and developers access to Praxis-1’s weights and fine-tuning code, leaving future evaluations to test its performance on additional tasks and physical systems.
For now, the clearest next steps are more evidence: independent comparisons using consistent setups, results on physical robots, and details from OpenAI about how its reorganized teams will pursue robotics. Until those emerge, the available reports show a change in organizational structure and a new specialist model—not a settled answer on which approach will lead.
Key Questions
What changed in OpenAI’s robotics organization?
According to Fortune, OpenAI moved its robotics researchers into the main research organization under Mark Chen and its applied robotics and hardware teams into the Applied organization under Greg Brockman.
Has OpenAI said it is leaving robotics?
No. The company’s spokesperson told Fortune that robotics remains part of OpenAI’s long-term vision. The reorganization changes where teams sit; the supplied reporting does not say OpenAI has ended its robotics work.
What is Runway’s Praxis-1?
Praxis-1 is a 27-billion-parameter model for robot action. Runway released its weights and fine-tuning code under the Apache 2.0 license, which permits commercial use.
Which model performed better on RoboLab-120?
The supplied figures report GPT-6 Astra at 74.8% in an agentic simulation setup and Praxis-1 at 46.2%. The results use different approaches and do not establish which model performs better on physical robots.
When might OpenAI show its humanoid robot?
CEO Sam Altman has targeted a first public demonstration in 2027, according to the source material. Further details about the timing and what the demonstration will include are not provided.
Source: ThorstenMeyerAI.com
